Such a ballast circuit is described in U.S. Pat. No. 5,723,953.
For igniting a (compact) fluorescent lamp two methods can be applied: warm start or cold start. Warm start means that during a specific time the electrodes are pre-heated while maintaining the lamp voltage below its ignition voltage. Because of the high electrode temperature the electrodes will start emitting electrons. When after pre-heating a lamp voltage is applied which is larger than the ignition voltage, an avalanche will take place and the fluorescent lamp will be on. As electrons are already available, the voltage across the electrodes is low during and after the ignition process, so no sputtering of the electrodes will occur, that is reflected in a high switching lifetime of the lamp.
For a cold start a high voltage is applied across the fluorescent lamp in a very short time. After ignition the electrodes needs to supply electrons to establish a lamp current. However the electrodes are cold and the only manner to generate electrons is to force them out of the electrode lattice by a high electric field. This high voltage will heat the electrode and eventually thermal emission will take place. In the time frame wherein the electrode voltage is high, the so called glow phase, sputtering of the electrode will take place that is reflected in a relatively short switching lifetime. The lamp driver should take care that the time frame wherein the electrode voltage is high is as short as possible. This means that in the glow phase maximum power should be delivered to the lamp to heat up the electrodes as quickly as possible. The lamp voltage during the glow phase is high (typically 500V, depending on the lamp type). The reason to apply the cold start mechanism is to minimize the costs of the ballast.
A lamp driver usually consists of a half-bridge topology. The pre-heating, ignition and burning states are obtained by sweeping down the frequency of the half-bridge switches over the resonance curve of the resonance (LC) network. The resonance frequency (1/(2*pi*sqrt(Llamp*Clamp)) often is chosen near the start frequency (the maximum frequency) for lowest current stress during ignition. Sweeping is often established via a voltage controlled oscillator.
For cold start the frequency sweep from the start frequency to the nominal frequency (the minimum frequency) is very short, for instance 10 ms, with respect to the glow time, which is more than 100 ms. Therefore without any measures the half-bridge will run at nominal frequency during the glow phase. Because this nominal frequency is far below the resonance frequency the half-bridge is not capable anymore of generating the high glow voltage, and furthermore the half bridge is also running in capacitive mode. As a result the lamp may extinguish, or alternatively remain in glow mode and be destroyed thereby. Known cold start lamp drivers which address this problem have non-integrated self-oscillating circuits which direct itself to resonance and thereby maximum power to the lamp in the glow phase. These circuits are however expensive and hard to integrate in an IC.